Add to_dataframe() to Record.

[Ivorforce]

Convenient for a quick start with records.

[tompollard]

Looks good to me, but I'll leave @bemoody and/or @cx1111 to comment too.

[tompollard]

Should there be an option to set data to different signal types, e.g. d_signal?

[Ivorforce]

Perhaps; I added a variable check to support all 4 combinations of data. I'm not sure if it makes sense for digital data though so I'll let someone else decide whether we want to keep it.

[cx1111]

It doesn't seem very efficient to store a regularly sampled signal as a dataframe. What's the use-case for this? To filter samples based on a datetime range? Perhaps it wold be better to add methods that get you the first index of the signal before, after, or between datetimes, which would incidentally solve for the d_signal/p_signal issue mentioned above.

Let me know if the dataframe is more useful for other reasons.

[tompollard]

It doesn't seem very efficient to store a regularly sampled signal as a dataframe. What's the use-case for this?

Maybe not efficient, but lots of people are familiar with Pandas DataFrames which is good for accessibility. This small chunk of code allows people to quickly play around with the data in a familiar structure. e.g. perhaps not a great example of a use case, but I could do something like:

View the data:

import wfdb
import matplotlib.pyplot as plt
import seaborn as sns

record = wfdb.rdrecord('sample-data/a103l')
df = record.to_dataframe()
df.head()

                              II         V     PLETH
0 days 00:00:00        -0.023596  0.867586  0.482203
0 days 00:00:00.004000 -0.036981  0.982985  0.544374
0 days 00:00:00.008000 -0.062923  0.859791  0.478212
0 days 00:00:00.012000 -0.092452  0.788783  0.442857
0 days 00:00:00.016000 -0.094522  0.851996  0.474302

Plot the distribution of measurements:

sns.boxplot(data=df)
plt.show()

[cx1111]

Ok, makes sense. @Ivorforce can you please add a unit test and an example in the demo notebook?

[Ivorforce]

Yes, sorry for not providing motivation - @tompollard got it exactly right. I opened this PR for accessibility to new people and compatibility to existing libraries.

Editing an ipython notebook is a bit of a struggle if you don't want to change metadata, but I added a unittest and a demo like asked. I also added support for digital and expanded signals. Let me know if it's ready like this.

The demo, for quick reference:

[bemoody]

This looks nice! I don't really use Pandas so I don't have much to add here. One question that comes to mind: is there a way for the dataframe to indicate the physical units of each channel/column?

[Ivorforce]

No support for units in dataframes. See this issue: pandas-dev/pandas#10349

[Ivorforce]

I switched to pd.Timedelta() because it has a higher precision than datetime.timedelta, resulting in a lower probability of precision loss for the index.

[tompollard]

@Ivorforce it looks like one of the tests is failing. Please could you take a look when you have time? See: https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true

_________________________ TestRecord.test_to_dataframe _________________________
[18](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:19)
[19](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:20)
self = <tests.test_record.TestRecord testMethod=test_to_dataframe>
[20](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:21)
[21](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:22)
    def test_to_dataframe(self):
[22](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:23)
        record = wfdb.rdrecord("sample-data/test01_00s")
[23](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:24)
        df = record.to_dataframe()
[24](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:25)
[25](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:26)
        self.assertEqual(record.sig_name, list(df.columns))
[26](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:27)
        self.assertEqual(len(df), record.sig_len)
[27](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:28)
>       self.assertEqual(df.index[0], pd.Timedelta())
[28](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:29)
[29](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:30)
tests/test_record.py:538: 
[30](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:31)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
[31](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:32)
[32](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:33)
>   ???
[33](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:34)
E   ValueError: cannot construct a Timedelta without a value/unit or descriptive keywords (days,seconds....)
[34](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:35)
[35](https://github.com/MIT-LCP/wfdb-python/runs/6904057193?check_suite_focus=true#step:5:36)
pandas/_libs/tslibs/timedeltas.pyx:986: ValueError

[Ivorforce]

Whoops, looks like I missed this. Tests should pass now.

[cx1111]

Ty!